FELGA: Unsupervised Fragment Embedding for Fine-Grained Cross-Modal Association | IEEE Conference Publication | IEEE Xplore